Rolling wrote on 2026-04-16, 09:44:

I haven't replaced any caps yet. The PSU is a be quiet! Blackline BQT P4-450W.

Might be worthwhile to open and check what's inside it... and maybe post some pictures here too (that is, if you prefer doing this first over getting another PSU.)
Any time I see a PSU that has multiple models (and thus multiple power ratings) on the label, I can almost guarantee you it's going to be a budget PSU at best. Just think about it: if the manufacturer is so cheap that they can't be bothered to make different labels for the different power ratings of that model line, do you think they bothered to actually make even a semi-decent PSU? I mean, it's not a given, but more often that not, this is true.

RetroPCCupboard wrote on 2026-04-17, 05:35:

So, personally, I think even if the caps are bad, it should still work.

I suggest not to go around and give such advice.

If you didn't design the CPU VRM circuit yourself or at least analyzed it deep enough to know for sure that it will be OK even with failing caps, then don't assume things will be OK just because you had that one experience where the bad caps didn't seem to affect the circuit. FWIW, failing caps won't heal or get better with time, so it will only be a matter of time before they go too out-of-spec for things to stop working.
I have seen a good deal of VRMs get further damage from the bad caps (e.g. burned MOSFET). Hard to put it in numbers, since it's very specific to each design - e.g. XFX GeForce 8x00 series with the terrible Sacon FZ tend to not care too much, even when these caps go completely open-circuit. But a similar era eVGA 7600 GT has a good 25-33% chance of blowing up a RAM chip. And old WRT54G routers? - depending on the internal version/design, some often got killed when the caps started failing.
So again, it's not a good idea to assume that things will be OK if you don't take the time to understand the specific VRM design. Some will be tolerant and/or have redundancy, and others not so much.

rasz_pl wrote on 2026-04-17, 00:17:

Those Caps do indeed look lovely and Asus used great capacitors.
I would start with another PSU. Then reluctantly look into caps situation. You can start by severely underclocking CPU.

ASUS did *not* always use good caps, though they did try to use good Japanese capacitor brands at least most of the time and occasionally would dip down to 2nd tier / Taiwanese brands that weren't great but weren't straight garbage either (e.g. OST and Ltec.)

With that said, it's never a good idea to assume that just because one has an ASUS board, it will come with good caps.

I do agree that it's probably best to start with another PSU, just so we could rule that out too... which would be tricky, since it needs to be something with a strong 5V rail. I personally like the standard ATX 250-300 Watt PSUs from Dell machines - typically the Delta's and HiPro's/Chicony's. Both of these have very solid 5V rails. The problem is they too tend to randomly suffer from bad caps. But even for people who don't have too much expertise in recapping, I think it would still be just as easy to buy and recap one of thees PSUs than to look for some unobtanium PSU with really high 5V rail rating... not to mention possibly much cheaper.

Shponglefan wrote on 2026-04-17, 10:41:

KZG series caps are notorious for failing, so those should definitely be replaced.

Concur.

The 6.3V 3300 uF are some of the worst from the KZG series when it comes to guaranteed failure, possibly only behind the 6.3V 820 uF KZG - at least in my experience of recapping old boards for about 15 years now. Only the 16V 1500 uF KZG seem to not have problems... or not yet.

rasz_pl wrote on 2026-04-17, 12:59:

if crashes get better after underclocking then its either cooling or cpu supply aka caps

Yes... or the PSU still.
In fact, the caps in the PSU don't have to be bad for the PSU to be producing out-of-spec ripple. I had this happen with several budget PSUs, where they simply didn't have enough filtering on their output rails. With lower-power systems, everything was stable as a rock. But move up to system with higher power draw, and random crashes/BSODs would manifest - on some systems randomly, and on others in very specific load cases.

So I don't think we are quite there yet to pinpoint exactly what component is causing the issue here.

If O/P has another (preferably old) PSU with a strong 5V rail and caps looking OK, I would try that first.
Alternatively, if O/P has a lower-power socket A CPU (e.g. a Duron or lower-end Athlon XP with Thoroughbred core), that should be tested too. This will rule out the CPU in general being a problem (be it from running too hot or the AXP 3200+ possibly being unstable.)
And if none of that seems to help, then the motherboard might be due for a recap.

Depending on availability of parts and/or skill level, the above three can be tackled in any order the O/P prefers.

Rolling wrote on 2026-04-17, 11:00:

Probably a good idea. Any hints where i could get a nice cap kit in europe?

Asides from the usual reputable online parts retailers (e.g. Mouser, Digikey, Farnel, and etc.), wasn't there also a sort of "local" parts retailer in Germany? IIRC, there was in Austria, but I forget what it was called. They didn't have that many choices, but were generally OK.
That said, even eBay, AliExpress, and Amazon might be OK for sourcing caps, so long as you actually get solid polymers instead of electrolytics... and you might have to, since ultra-low ESR electrolytic caps are not really made any more. The closest you can get to United Chemicon KZG (if you go with electrolytic caps again) is probably Chemicon KZM or Panasonic FS or Nichicon HW or Rubycon ZLH/ZLK... all of which are slightly worse (not as well-rated) than KZG, so you may have to over-spec on capacitance/can size a little.

*edit*
As @shevalier suggested in the previous post, Wurth polymer caps could be a good alternative if you can get them locally.

If you figure out which are the CPU "low" side caps (i.e CPU V_core), then for those you can go with 1500-2200 uF polymers with 2.5V or 4V ratings.
For the "high" side caps (5V rail from the PSU), you can probably drop down further in capacitance (e.g. 820 to 1200 uF), but voltage rating must be at least 6.3V as the original KZGs.

Now, I don't know if I had traced what's what on socket A Asus boards before, but my hunch is that these 5 KZG are the CPU VRM "low" side:
download/file.php?id=240221&mode=view

And these three KZG should be the "high" side (e.g. where you need 6.3V or 10V -rated caps):
download/file.php?id=240220&mode=view

If someone has this board and the chance to double-check me, please do.

*edit 2*
I just looked at pictures of this board on TRW, an I think the above should be correct.

Also, have a look at the smaller OST caps in the upper-right corner of the board by the RAM slots, since there is another buck VRM there. A possibly 3rd buck VRM is right under the AGP slot. You can tell these by the inductors. Buck VRM + bad caps almost always results in instability and/or crashes.
Generally the older OST (RLP series on this board?) are a little more stable than their newer stuff, but I always try to double-check them whenever I can. The thing about OST is that they sometimes don't bulge or show any visual signs at all when they start to fail - particularly the RLP series.

momaka wrote on 2026-04-17, 14:46:

I suggest not to go around and give such advice.

If you didn't design the CPU VRM circuit yourself or at least analyzed it deep enough to know for sure that it will be OK even with failing caps, then don't assume things will be OK just because you had that one experience where the bad caps didn't seem to affect the circuit.

Fair enough. I wouldn't have said that as a blanket statement. I know that bad caps cause problems. Just saying that my Asus A7n8X had bad caps and worked, so presumably the circuit is at least a bit tolerant of this. I did say in a later response that the errors did look like a hardware problem. That could include caps.

momaka wrote on 2026-04-17, 14:46:

FWIW, failing caps won't heal or get better with time, so it will only be a matter of time before they go too out-of-spec for things to stop working.
I have seen a good deal of VRMs get further damage from the bad caps (e.g. burned MOSFET). Hard to put it in numbers, since it's very specific to each design - e.g. XFX GeForce 8x00 series with the terrible Sacon FZ tend to not care too much, even when these caps go completely open-circuit. But a similar era eVGA 7600 GT has a good 25-33% chance of blowing up a RAM chip. And old WRT54G routers? - depending on the internal version/design, some often got killed when the caps started failing.
So again, it's not a good idea to assume that things will be OK if you don't take the time to understand the specific VRM design. Some will be tolerant and/or have redundancy, and others not so much.

Noted. Thanks for the detailed response.

I think I have read somewhere that the nForce boards are more prone to failure than the alternative Socket A chipsets. Am I correct on that, or am I misremembering? I seem to remember a PhilsComputerLab video also where he had several of these boards that were bad.

momaka wrote on 2026-04-17, 14:46:

*edit* As @shevalier suggested in the previous post, Wurth polymer caps could be a good alternative if you can get them locally. […]
Show full quote

*edit*
As @shevalier suggested in the previous post, Wurth polymer caps could be a good alternative if you can get them locally.

If you figure out which are the CPU "low" side caps (i.e CPU V_core), then for those you can go with 1500-2200 uF polymers with 2.5V or 4V ratings.
For the "high" side caps (5V rail from the PSU), you can probably drop down further in capacitance (e.g. 820 to 1200 uF), but voltage rating must be at least 6.3V as the original KZGs.

To be honest, for loads of this magnitude, polymer capacitors from AliExpress sold by reputable sellers are more than adequate.
Capxon (now part of Teapo), Lelon or Samwha are ideal.
Würst is roughly on a par with them (and that’s a very high standard).

The main challenge is drawing up a shopping list.
The capacitors must have exactly the same distance between their terminals as the original ones.
Voltages
For the CPU – 3.3V, for the 3.3V bus – 4 or 6.3V, for the 5V bus – 6.3V, for 12V – 16V.
Capacitance – from the original value down to 1/3 of it.
Then you need to minimise the list.
Do not buy (for example) 2 x 470μF 3.3V capacitors and 12 x 560μF 6.3V capacitors; instead, buy 14 of the latter straight away.

An A4 sheet of paper with the capacitors drawn on it and a multimeter make life much easier.

Thanks for all the ideas and general advice on how to deal with these kinds of issues. Really great information, i'm learning a lot!

I found another PSU from Delux with 30A on the 5V and 16A on the 12 rail. I tested the system by just connecting the ATX plug, one HDD and the Radeon x850 via Molex. I measured the voltage on one molex plug, it never drops under 12,27.
=> Test result: Same effects as with the be quiet PSU.

I also found a Athlon XP 2000+ running at 1,6 GHz (133 FSB) and tried running the system with this CPU.
=> Test result: Same effects as with the Athlon 3200+.

At this point it's probably (?) safe to assume, that the caps gone bad. I'll try to get the parts and will replace them.

momaka wrote on 2026-04-17, 14:46:

Concur.

The 6.3V 3300 uF are some of the worst from the KZG series when it comes to guaranteed failure, possibly only behind the 6.3V 820 uF KZG - at least in my experience of recapping old boards for about 15 years now. Only the 16V 1500 uF KZG seem to not have problems... or not yet.

All values of the older KZG series (2005 and earlier) have the problem, be it 16v 1500, or any capacitance whatsoever.

Success! The caps were really the problem. I decided to change the big ones in the CPU area (8x Kemet A750 2.200µF 16V) and a lot of smaller ones on the whole board (16x Kemet A750 1.000µF 6,3V).

Since the swap not a single crash and everything seems to run just fine. I think the system turned out quite well in the end.

Thanks for all your support & the detailed information in the process.

Rolling wrote on 2026-04-22, 15:19:

Success! The caps were really the problem. I decided to change the big ones in the CPU area (8x Kemet A750 2.200µF 16V) and a lot of smaller ones on the whole board (16x Kemet A750 1.000µF 6,3V).
...
Since the swap not a single crash and everything seems to run just fine. I think the system turned out quite well in the end.

Awesome and congrats on getting the board fixed! 😀

PcBytes wrote on 2026-04-17, 16:40:

All values of the older KZG series (2005 and earlier) have the problem, be it 16v 1500, or any capacitance whatsoever.

Well, you might be right.
Looking at my used caps stock, I don't really have that many 16V 1500 uF KZG from before 2005. Almost all of them are from 2006 and later. I think I have only 3 or so specimens from an older (circa 2003/4) motherboard... but I don't remember where I placed these.